fix(sync): retry failed screenshot fetches instead of caching forever

[miaulalala]

Summary

• Investigated App Store usercontent proxy returns "File not found" for valid CoBudget screenshots appstore#1843 (CoBudget screenshots showing "File not found" through the proxy while working fine directly on GitHub).
• Found that sync.php writes the failure placeholder (generateWarningImage()) into the exact same cache path that file_exists($cacheUrl) uses to decide "already handled, skip forever" — so any transient fetch failure (timeout, temporary origin error, unrecognized mime type) becomes permanent and is never retried.
• Also found that a single uncaught error while processing one screenshot (e.g. mime_content_type() returning false, which then fails str_starts_with()'s type check) aborts the entire sync run, leaving every app later in the list unsynced until the same bug is hit again next run.
• This PR adds a cache-failed/ marker directory that tracks failed URLs with a 6h backoff, so failures get retried instead of being stuck forever, and wraps per-screenshot handling in handleApps() so one bad entry can't kill the whole run.

Test plan

• php -l sync.php passes
• Manually exercised the extracted handleScreenshot() logic against a real failing URL: confirmed a fail marker is created on failure, a retry within the backoff window is a no-op, and a retry after the backoff window elapses re-attempts the fetch
• Confirmed a subsequent successful fetch clears the fail marker and replaces the cached placeholder with the real image
• Would appreciate a maintainer sanity-check against the production cache layout / cron invocation before merge

🤖 Generated with Claude Code

[edward-ly]

🚀